You’re not buying intelligence. You’re paying for the right to teach it.

I spent years inside Google watching how platforms actually work. Not the feature set — the economics underneath it. The pattern doesn’t vary: you bring the intent, the signal, the behavior. The platform captures it, generalizes it at scale, and sells the insight back to whoever will pay for it.

AI is running the same play. Alex Karp said the quiet part loud enough: the product often isn’t the answer you receive — it’s the data you hand over to get it. Every workflow, every strategy memo, every customer signal you route through an external frontier model is curriculum. The lab internalizes your specificity. You get useful outputs. They get something you can’t take back: generalized value built from your particular patterns.

In practice, that looks like buying back capability you effectively taught the model. I’ve watched platform dynamics long enough to call it what it is.

The thought experiment about pricing AI at 30% of the upside is useful as a provocation — it names the right thing (value participation, not token throughput) — and useless as a model. Three reasons.

Value is probabilistic and multi-causal. AI almost never delivers a single, traceable outcome you can audit. Attribution would be a legal nightmare. And competitive pressure plus infrastructure economics push every lab toward token and cloud-style pricing regardless of the value delivered.

So the percentage debate is a distraction. The real question is the one I call govern — the fourth dimension in the AI Fluency framework I use with every leadership team I work with: who owns what you’re feeding in?

What happens to your data after the session ends? What guarantees exist on usage and retention — in writing, not buried in terms-of-service? Are you treating this as a tool, or as a strategic partnership you’ve actually negotiated?

Here’s what I see consistently: most leadership teams are waiting until they fully understand the technology before they set the rules around it. That’s the wrong order. You don’t need to be an AI expert to govern how your organization uses one. Waiting for fluency before acting is exactly how you lose the thing that makes you worth working with in the first place.

If you’re running anything that constitutes genuine IP through an external frontier model — proprietary frameworks, competitive intelligence, client behavior patterns — you’re already in that partnership. The question is whether you’ve acknowledged it.

Name the posture. Then set the terms.